| output_dir: "logs/s1" | |
| train: | |
| seed: 1234 | |
| epochs: 15 | |
| batch_size: 8 | |
| save_every_n_epoch: 5 | |
| precision: 32 | |
| if_save_latest: true | |
| if_save_every_weights: true | |
| exp_name: "gpt_training" | |
| half_weights_save_dir: "weights/s1" | |
| wandb: | |
| project: "gpt-sovits-hindi" | |
| name: "stage1_training" | |
| entity: null | |
| log_interval: 100 | |
| optimizer: | |
| lr_init: 0.0001 | |
| lr: 0.0004 | |
| lr_end: 0.00001 | |
| warmup_steps: 500 | |
| decay_steps: 1000 | |
| data: | |
| training_files: "data8" | |
| max_sec: 60 | |
| max_frames: 60 | |
| filter_length: 2048 | |
| hop_length: 640 | |
| win_length: 2048 | |
| mel_channels: 128 | |
| mel_fmin: 0.0 | |
| mel_fmax: null | |
| cleaned_text: true | |
| num_workers: 4 | |
| batch_size: 8 | |
| pad_val: 1024 | |
| # Data paths | |
| train_semantic_path: "data8/semantic.tsv" | |
| train_phoneme_path: "data8/phoneme.txt" | |
| model: | |
| hidden_dim: 768 | |
| embedding_dim: 768 | |
| n_layer: 12 | |
| head: 12 | |
| n_embd: 768 | |
| vocab_size: 2048 | |
| block_size: 1000 | |
| embd_pdrop: 0.1 | |
| resid_pdrop: 0.1 | |
| attn_pdrop: 0.1 | |
| semantic_dim: 1024 | |
| num_layers: 6 | |
| ffn_hidden: 3072 | |
| dropout: 0.1 | |
| attention_dropout: 0.1 | |
| hidden_dropout: 0.1 | |
| max_text_positions: 2048 | |
| max_mel_positions: 8000 | |
| prenet_dim: 384 | |
| postnet_dim: 384 | |
| prenet_layers: 3 | |
| postnet_layers: 3 | |
| phoneme_vocab_size: 2048 | |
| EOS: 2047 | |
| pad_val: 1024 |